How To Disassemble A Windows Program

After we've found and analyzed WinMain() (-> lesson 1), the next

places to inspect when you crack a program are the windows procedures

and dialog procedures (this is true only for Windows *programs*; for

DLL, on the countrary, the cracking procedures are different and the

relvant techniques will be discussed in another lesson).

These WndProcs and DialogProcs are "callback" procedures: they are

*exported* from Windows executables, almost as the program were a DLL,

so that Windows can call them.

And -hear, hear!- beacuse they are exported these crucial procedures

have *names* (almost always useful) that are accessible to any decent

Windows disassembler. In Taskman.lst, for example, WCB clearly

identifies TASKMANDLGPROC:

Exported names by location:

1:007B 1 TASKMANDLGPROC <- It's a DialogProc !

It works out well that the WndProcs and DialogProcs show up so nicely

in the disassembled listings, because, as we know from Windows

programming, these subroutines are "where the action is" in event

driven Windows applications... or at least where the action begins.

Furthermore we know that these subroutines will be most likely little

more than (possibly very large) message handling switch/case

statements. These usually look something like this: long FAR PASCAL

_export WndProc(HWND hWnd, WORD message, WORD wParam, LONG lPAram)

long FAR PASCAL _export WndProc(HWND hWnd, WORD message, WORD

wParam, LONG lPAram)

{ ...

switch (message)

{

case WM_CREATE:

//... handle WM_CREATE message

break;

case WM_COMMAND:

//... handle WM_COMMAND message

break;

default:

return DefWindowProc(hwnd, message, wParam, lParam);

}

}

Wow! Yes! As you already guessed this means that... that we get

immediately 4 parameters for EACH exported WndProc or DlgProc!

Actually there's no rule that states that a Windows WndProc or DlgProc

has to look like this... it's just that they almost always do!

Here is how the parameters to the WndProc or DialogProc will appear in

the assembly language listing (after the function prologue):

long FAR PASCAL _export WndOrDialogProc(HWND hwnd, WORD

message, WORD wParam, LONG lParam);

lParam = dword ptr [bp+6]

wParam = word ptr [bp+0Ah]

message = word ptr [bp+0Ch]

hWnd or hWndDlg = word ptr [bp+0Eh]

With this knowledge, we can replace an otherwise meaningless [bp+0Ch]

with a label such as "message", a [bp+0Eh] with a "hwnd" or "hwndDlg",

and so on in *ANY* DialogProc and WndProc in *ANY* Windows program.

The boilerplate nature of Windows programming greatly simplifies

cracking. For example, here is part of our Taskman exported:

The problem here, of course, is what to make of all these magic

numbers: 0064, OO1C, 00F4 and so on... how are we going to figure out

what these mean?

DialogProc: TASKMANDLGPROC:

1.007B ; TASKMANDLGPROC

... (function prologue)

1.008A 8B760E mov si, hWndDlg ;[bp+0E]

1.008D 56 push si

1.008E 6A64 push 0064

1.0090 9AFFFF0000 call USER.GETDLGITEM

1.0095 8BF8 mov di, ax

1.0097 8B460C mov ax, message ;[bp+0C]

1.009A 2D1C00 sub ax, 001C

1.009D 7416 je 00B5

1.009F 2DF400 sub ax, 00F4

1.00A2 7436 je 00DA

1.00A4 48 dec ax

1.00A5 7503 jne 00AA

1.00A7 E98301 jmp 022D

1.00AA >2D5303 sub ax, 0353

1.00AD 7503 jne 00B2

1.00AF E9D602 jmp 0388

1.00B2 >E9C801 jmp 027D

1.00B5 >837E0A00 cmp word ptr wParam, 0 ;[bp+0A]

1.00B9 7403 je 00BE

1.00BB E9BF01 jmp 027D

...

When examined via disassembled listings, Windows programs tend to

contain a lot of "magic numbers". Of course the actual source code

would be :

* #include '<'windows.h'>' and

* #define numeric constants for the various resources (menus,

strings, dialog controls, etc.) that it uses.

Given a disassembled listing, it should be possible to turn a lot of

these seemingly senseless numbers back into something understandable.

Let's start with the number 001C in TaskManDlgProc():

1.0097 8B460C mov ax, message ;[bp+0C]

1.009A 2D1C00 sub ax, 001C

1.009D 7416 je 00B5

If AX holds the *message* parameter to TaskManDlgProc() (line

1.0097)... then the value 001C must be a Windows WM_ message number

(one of those you can breakpoint to with WINICE's BMSG command, by the

way). Looking in WINDOWS.H, we find that 0x1C is WM_ACTIVATEAPP.

TaskManDlgProc() is subtracting this value from AX and then jumping

somewhere (let's call it ON_ACTIVATEAPP) if the result is zero... i.e.

if it is WM_ACTIVATEAPP.

This is an odd way to test whether (message == WM_ACTIVATEAPP): if the

test fails, and we do not take the jump to ON_ACTIVATEAPP, the message

number has 1C subtracted from it... and this value must be taken

account of by the next switch statement:

1.009F 2DF400 sub ax, 00F4 ; (+1C=110=WM_INITDIALOG)

1.00A2 7436 je 00DA ; jump to ON_INITDIALOG

1.00A4 48 dec ax ; (110+1=111=WM_COMMAND)

1.00A5 7503 jne 00AA ; no, go elsewhere

1.00A7 E98301 jmp 022D ; yes, jump to ON_COMMAND

Other WndProcs & DialogProcs will contain straightforward tests,

rather than testing via subtraction... is a matter of compiler choice.

In any case, a WndProc or DialogProc generally contains a collection

of handlers for different messages.

In the case of TaskManDlgProc(), we can see that's handling

WM_ACTIVATEAPP, WM_INITDIALOG and WM_COMMAND. By itself, this

information is rather boring... however, it tells us what is happening

*elsewhere* in the function: 1.00B5 must be handling WM_ACTIVATEAPP

messages (therefore let's call it ON_ACTIVATEAPP), 1.00DA must be

handling WM_INITDIALOG, and 1.022D must be handling WM_COMMAND

messages.

Write it down! This same basic technique -find where the [bp+0Ch]

"message" parameter to the WndProc or DialogProc is being rested, and

from that identify the locations that handle various messages- can be

used in *ANY* Windows program.

Because handling messages is mostly what Windows applications do, once

we know where the message handling is, we pretty much can have our way

with the disassembled listing.

Let's look now at TaskManDlgProc():

TASKMANDLGPROC proc far

...

DISPATCH_ON_MSG:

1.0097 8B460C mov ax, message ;[bp+0C]

1.009A 2D1C00 sub ax, WM_ACTIVATEAPP ;001C

1.009D 7416 je ON_ACTIVATEAPP

1.009F 2DF400 sub ax, 00F4 ; (+1C=110=WM_INITDIALOG)

1.00A2 7436 je ON_INITDIALOG

1.00A4 48 dec ax ;(110+1=111=WM_COMMAND)

1.00A5 7503 jne DEFAULT

1.00A7 E98301 jmp ON_COMMAND

DEFAULT:

1.00AA >2D5303 sub ax, 0353 ;(111+353=464=WM_USER+64

1.00AD 7503 jne ON_PRIVATEMSG ;00B2= some private msg

1.00AF E9D602 jmp 0388

ON_PRIVATEMSG:

1.00B2 >E9C801 jmp 027D

ON_ACTIVATEAPP:

1.00B5 >837E0A00 cmp word ptr wParam, 0 ;[bp+0A]

... ; code to handle WM_ACTIVATEAPP

ON_INITDIALOG:

... ; code to handle WM_INITDIALOG

ON_COMMAND:

... ; code to handle WM_COMMAND

1.022D >8B460A mov ax, wParam ;[bp+0A]

1.0230 3D6800 cmp ax, 0068 ; ? What's this ?

1.0233 7503 jne 0238

1.0235 E93301 jmp 036B

...

This is starting to look pretty reasonable. In particular, once we

know where WM_COMMAND is being handled, we are well on the way to

understand what the application does.

WM_COMMAND is *very* important for understanding an application

behavior because the handler for WM_COMMAND is where it deals with

user commands such as Menu selections and dialog push button clicks...

a lot of what makes an application unique.

If you click on "Cascade" in Task manager, for instance, it comes as a

WM_COMMAND, the same occurs if you click on "Tile" or "Switch To" or

"End Task".

An application can tell which command a user has given it by looking

in the wParam parameter to the WM_COMMAND message.

This is what we started to see at the ned of the TaskManDlgProc()

exerpt:

; We are handling WM_COMMAND, therefore wParam is here idItem,

; i.e. a control or menu item identifier

1.022D >8B460A mov ax, wParam ;[bp+0A]

1.0230 3D6800 cmp ax, 0068 ;ID number for a dialog control

1.0233 7503 jne 0238

1.0235 E93301 jmp 036B

1.0238 >7603 jbe 023D

1.023A E96001 jmp 039D

1.023D >FEC8 dec al ;1

1.023F 7420 je 0261 ;if wParam==1 goto 1.0261

1.0241 FEC8 dec al ;1+1=2

1.0243 7503 jne 0248

1.0245 E94701 jmp 038F ;if wParam==2 goto 1.038F

1.0248 >2C62 sub al, 62 ;2+62=64

1.024A 742A je 0276

1.024C FEC8 dec al ;64+1=65

1.024E 7432 je 0282

1.0250 2C01 sub al, 01 ;65+1=66

1.0252 7303 jnb 0257

1.0254 E94601 jmp 039D

1.0257 >2C01 sub al, 01 ;66+1=67

1.0259 7703 ja 025E

1.025B E9D200 jmp 0330

It's clear that wParam is being compared (in an odd subtraction way)

to valus 1,2,65,66 and 67. What's going on?

The values 1 and 2 are standard dialog button IDs:

#define IDOK 1

#define IDCANCEL 2

Therefore we have here the two "classical" push buttons:

1.023D >FEC8 dec al ; 1 = OK

1.023F 7420 je ON_OK ; If 1 goto 1.0261= ON_OK

1.0241 FEC8 dec al ; 1+1=2= CANCEL

1.0243 7503 jne NOPE ; goto neither OK nor CANCEL

1.0245 E94701 jmp ON_CANCEL ; if 2 goto 1.038F= ON_CANCEL

The numbers 65, 66 etc are specific to TaskManager however, we will

not find them inside WINDOWS.H... so there is no home to find the

names of the commands to which these magic number correspond, unless

we happen to have a debug version of the program true? NO! FALSE!

One of the notable things about Windows is that remarkably little

information is lost or thrown away compiling the source code. These

magic numbers seem to correspond in some way to the different Task

Manager push buttons... it's pretty obvious that there must be a way

of having applications tell Windows what wParam they want sent when

one of their buttons is clicked or when one of their menu items is

selected.

Applications almost always provide Windows with this information in

their resources (they could actually define menus and controls

dynamycally, on the fly, but few applications take advantage of this).

These resources are part of the NE executable and are available for

our merry snooping around.

This inspections of the resources in an EXE file is carried out by

means of special utilities, like RESDUMP, included with Windows source

(-> in my tool page). For example (I am using "-verbose" mode):

DIALOG 10 (0Ah), "Task List" [ 30, 22,160,107]

FONT "Helv"

LISTBOX 100 (64h), "" [ 3, 3,154, 63]

DEFPUSHBUTTON 1 (01h), "&Switch To" [ 1, 70, 45, 14]

PUSHBUTTON 101 (65h), "&End Task" [ 52, 70, 45, 14]

PUSHBUTTON 2 (02h), "Cancel" [103, 70, 55, 14]

STATIC 99 (63h), "" [ 0, 87,160, 1]

PUSHBUTTON 102 (66h), "&Cascade" [ 1, 90, 45, 14]

PUSHBUTTON 103 (67h), "&Tile" [ 52, 90, 45, 14]

PUSHBUTTON 104 (68h), "&Arrange Icons" [103, 90, 55, 14]

YEAH! It's now apparent what the numbers 64h, 65h etc. mean. Imagine

you would write Taskmanager yourself... you would write something on

these lines:

#define IDD_SWITCHTO IDOK

#define IDD_TASKLIST 0x64

#define IDD_ENDTASK 0x65

#define IDD_CASCADE 0x66

#define IDD_TILE 0x67

#define IDD_ARRANGEICONS 0x68

Let's look back at the last block of code... it makes now a lot more

sense:

ON_COMMAND:

; We are handling WM_COMMAND, therefore wParam is here idItem,

; i.e. a control or menu item identifier

1.022D >8B460A mov ax, wParam ;[bp+0A]

1.0230 3D6800 cmp ax, 0068 ;is it the ID 68h?

...

1.023D >FEC8 dec al ;1=IDOK=IDD_SWITCHTO

1.023F 7420 je ON_SWITCHTO ;0261

1.0241 FEC8 dec al ;1+1=2=ID_CANCEL

1.0243 7503 jne neither_OK_nor_CANCEL ;0248

1.0245 E94701 jmp ON_CANCEL ;038F

neither_OK_nor_CANCEL:

1.0248 >2C62 sub al, 62 ;2+62=64= IDD_TASKLIST

1.024A 742A je ON_TASKLIST ;0276

1.024C FEC8 dec al ;64+1=65= IDD_ENDTASK

1.024E 7432 je ON_ENDTASK ;0282

1.0250 2C01 sub al, 01 ;65+1=66= IDD_CASCADE

1.0252 7303 jnb check_for_TILE ;0257

1.0254 E94601 jmp 039D ;something different

check_for_TILE:

1.0257 >2C01 sub al, 01 ;66+1=67= IDD_TILE

1.0259 7703 ja 025E ;it's something else

1.025B E9D200 jmp ON_TILE_or_CASCADE ;0330

In this way we have identified location 0330 as the place where

Taskman's "Cascade" and "Tile" buttons are handled... we have renaimed

it ON_TILE_or_CASCADE... let's examine its code and ensure it makes

sense:

ON_TILE_or_CASCADE:

1.0330 >56 push hwndDlg ;si

1.0331 6A00 push 0000

1.0333 9A6F030000 call USER.SHOWWINDOW

1.0338 9A74030000 call USER.GETDESKTOPWINDOW

1.033D 8BF8 mov di, ax ;hDesktopWnd

1.033F 837E0A66 cmp word ptr wParam, 0066 ;IDD_CASCADE

1.0343 750A jne ON_TILE ;034F

1.0345 57 push di ;hDesktopWnd

1.0346 6A00 push 0000

1.0348 9AFFFF0000 call USER.CASCADECHILDWINDOWS

1.034D EB2F jmp 037E

ON_TILE:

1.034F >57 push di

1.0350 6A10 push 0010

1.0352 9AFFFF0000 call USER.GETKEYSTATE

1.0357 3D0080 cmp ax, 8000

1.035A 7205 jb 0361

1.035C B80100 mov ax, 0001 ;1= MDITILE_HORIZONTAL

1.035F EB02 jmp 0363

1.0361 >2BC0 sub ax, ax ;0= MDITILE_VERTICAL

1.0363 >50 push ax

1.0364 9AFFFF0000 call USER.TILECHILDWINDOWS

1.0369 EB13 jmp 037E

Yes, it makes a lot of sense: We have found that the "Cascade" option

in Tile manager, after switching through the usual bunch of

switch/case loops, finally ends up calling an undocumented Windows API

function: CascadeChildWindows()... similarly, the "Tile" routine ends

up calling TileChildWindow().

One thing screams for attention in the disassembled listing of

ON_TILE: the call to GetKeyState().

As an example of the kind of information you should be able to gather

for each of these functions, if you are serious about cracking, I'll

give you now here, in extenso, the definition from H. Schildt's

"General purpose API functions", Osborne's Windows Programming Series,

Vol. 2, 1994 edition (I found both this valuable book and its

companion: volume 3: "Special purpose API functions", in a second hand

shop, in february 1996, costing the equivalent of a pizza and a

beer!). Besides this function is also at times important for our

cracking purposes, and represents therefore a good choice. Here the

description from pag.385:

void GetKeyState(int iVirKey)

Use GetKeyState() to determine the up, down or toggled status of

the specified virtual key. iVirKey identifies the virtual key. To

return the status of a standard alphanumeric character in the

range A-Z, a-z or 0-9, iVirKey must be set equal to its ANSI

ASCII value. All other key must use their related virtual key

codes. The function returns a value indicating the status of the

selected key. If the high-order bit of the byte entry is 1, the

virtual key is pressed (down); otherwise it is up. If you examine

a byte emlement's low-order bit and find it to be 1, the virtual

key has been toggled. A low-order bit of 0 indicates that the key

is untoggled.

Under Windows NT/Win32, this function returns type SHORT.

Usage:

If your application needs to distinguish wich ALT, CTRL, or SHIFT

key (left or right) has been pressed, iVirKey can be set equal to

one of the following:

VK_LMENU VK_RMENU

VK_LCONTROL VK_RCONTROL

VK_LSHIFT VK_RSHIFT

Setting iVirKey equal to VK_MENU, VK_CONTROL or VK_SHIFT

instructs GetKeyState() to ignore left and right, and only to

report back the status of teh virtual key category. This ability

to distinguish among virtual-key states is only available with

GetKeyState() and the related functions listed below.

The following fragment obtains the state of the SHIFT key:

if(GetKeyState(VK_SHIFT) {

...

}

Related Functions:

GetAsyncKeyState(), GetKeyboardState(), MapVirtualKey(),

SetKeyboardState()

Ok, let's go on... so we have in our code a "funny" call to

GetKeyState(). Because the Windows USer's Guide says nothing about

holding down a "state" (shift/ctrl/alt) key while selecting a button,

this sounds like another undocumented "goodie" hidden inside TASKMAN.

Indeed, if you try it out on the 3.1 Taskman, you'll see that clicking

on the Tile button arranges all the windows on the desktop side by

side, but if you hold down the SHIFT key while clicking on the Tile

button, the windows are arranged in a stacked formation.

To summarize, when the 3.1. Taskman Tile button is selected, the code

that runs in response looks like this:

Tile:

ShowWindow(hWndDlg, SW_HIDE); // hide TASKMAN

hDesktopWnd = GetDesktopWindow();

if (GetKeyState(VK_SHIFT) == 0x8000)

TileChildWindows(hDesktopWnd, MDITILE_HORIZONTAL);

else

TileChildWindows(hDesktopWnd, MDITILE_VERTICAL);

Similarly, the CASCADE option in 3.1. TASKMAN runs the following code:

Cascade:

ShowWindow(hWndDlg, SW_HIDE); // hide TASKMAN

CAscadeChildWindows(GetDesktopWindow(), 0);

We can then proceed through each TASKMAN option like this, rendering

the assembly language listing into more concise C.

The first field to examine in TASKMAN is the Task List itself: how is

the "Task List" Listbox filled with the names of each running

application?

What the List box clearly shows is a title bar for each visible top

level window, and the title bar is undoubtedly supplied with a call to

GetWindowText()... a function that obtains a copy of the specified

window handle's title.

But how does TASKMAN enumerate all the top-level Windows? Taskman

exports TASKMANDLGPROC, but does not export any enumeration procedure.

Most of the time Windows programs iterate through all existing windows

by calling EnumWindows(). Usually they pass to this function a pointer

to an application-supplied enumeration function, which therefore MUST

be exported. This callback function must have following prototype:

BOOL CALLBACK EnumThreadCB(HWND hWnd, LPARAM lParam)

Of course, the name a programmer chooses for such an exported function

is arbitrary. hWnd will receive the handle of each thread-associated

window.lParam receives lAppData, a 32-bit user- defined value. This

exported function must return non-zero to receive the next enumerated

thread-based window, or zero to stop the process.

But here we DO NOT have something like TASKMANENUMPROC in the list of

exported functions... what's going on? Well... for a start TASKMAN IS

NOT calling EnumWindows()... Taskman uses a GetWindow() loop to fill

the "Task List" list box, study following C muster, sipping a good

cocktail and comparing it with the disassembled code you have printed:

Task List:

listbox = GetDlgItem(hWndDlg, IDD_TASKLIST);

hwnd = GetWindow(hwndDlg, GW_HWNDFIRST);

while (hwnd)

{ if ((hwnd != hwndDlg) && //excludes self from list

IsWindowVisible(hwnd) &&

GetWindow(hwnd, GW_OWNER))

{ char buf[0x50];

GetWindowText(hwnd, buf, 0x50); // get titlebar

SendMessage(listbox, LB_SETITEMDATA,

SendMessage(listbox, LB_ADDSTRING, 0, buf),

hwnd); // store hwnd as data to go

} // with the titlebar string

hwnd = GetWindow(hwnd, GW_HWNDNEXT);

}

SendMessage(lb, LB_SETCURSEL, 0, 0); // select first item

The "End Task" opton in Taskman just sends a WM_CLOSE message to the

selected window, but only if it's not a DOS box. TASKMAN uses the

undocumented IsWinOldApTask() function, in combination with the

documented GetWindowTask() function, to determine if a given HWND

corresponds to a DOS box:

End Task:

... // boring details omitted

if(IsWinOldApTask(GetWindowTask(hwndTarget)))

MaybeSwitchToSelecetedWindow(hwndTarget);

if(IsWindow(hwndTarget) &&

(! (GetWindowLong(hwndTarget, GWL 5STYLE) & WS_DISABLED))

{

PostMessage(hwndTarget, WM_CLOSE, 0, 0);

}

The "Arrange Icons" option simply runs the documented

ARrangeIconicWindows() function:

Arrange Icons:

Showwindow(hWndDlg, SW_HIDE);

ArrangeIconiCWindows(GetDesktopWindow());

The "Switch To" option in TASKMAN is also interesting. Like "Tile" and

"Cascade", this too it's just a user-interface covering an

undocupented Windows API function, in this case SwitchToThisWindow().

Let's walk through the process of deciphering a COMPLETELY unlabelled

Windows disassembly listing, that will be most of the time your

starting situation when you crack, and let's turn it into a labelled C

code.

By the way, there does exist an interesting school of research, that

attempts to produce an "EXE_TO_C" automatical converter. The only

cracked version of this program I am aware of is called E2C.EXE, is

198500 bytes long, has been developed in 1991 by "The Austin Code

Works and Polyglot International" in Jerusalem (Scott Guthery:

guthery@acw.com), and has been boldly brought to the cracking world by

Mithrandir/AlPhA/MeRCeNarY. Try to get a copy of this tool... it can

be rather interesting for our purposes ;-)

Here is the raw WCB disassembled code for a subroutine within TASKMAN,

called from the IDD_SWITCHTO handling code in TaskManDlgProc():

1.0010 >55 push bp

1.0011 8BEC mov bp, sp

1.0013 57 push di

1.0014 56 push si

1.0015 FF7604 push word ptr [bp+04]

1.0018 681A04 push 041A

1.001B FF7604 push word ptr [bp+04]

1.001E 680904 push 0409

1.0021 6A00 push 0000

1.0023 6A00 push 0000

1.0025 6A00 push 0000

1.0027 9A32000000 call USER.SENDMESSAGE

1.002C 50 push ax

1.002D 6A00 push 0000

1.002F 6A00 push 0000

1.0031 9AEF010000 call USER.SENDMESSAGE

1.0036 8BF8 mov di, ax

1.0038 57 push di

1.0039 9A4C000000 call USER.ISWINDOW

1.003E 0BC0 or ax, ax

1.0040 742A je 006C

1.0042 57 push di

1.0043 9AFFFF0000 call USER.GETLASTACTIVEPOPUP

1.0048 8BF0 mov si, ax

1.004A 56 push si

1.004B 9AA4020000 call USER.ISWINDOW

1.0050 0BC0 or ax, ax

1.0052 7418 je 006C

1.0054 56 push si

1.0055 6AF0 push FFF0

1.0057 9ACD020000 call USER.GETWINDOWLONG

1.005C F7C20008 test dx, 0800

1.0060 750A jne 006C

1.0062 56 push si

1.0063 6A01 push 0001

1.0065 9AFFFF0000 call USER.SWITCHTOTHISWINDOW

1.006A EB07 jmp 0073

1.006C >6A00 push 0000

1.006E 9ABC020000 call USER.MESSAGEBEEP

1.0073 >5E pop si

1.0074 5F pop di

1.0075 8BE5 mov sp, bp

1.0077 5D pop bp

1.0078 C20200 ret 0002

The RET 0002 at the end tells us that this is a near Pascal function

that expects one WORD parameter, which appears as [bp+4] at the top of

the code.

Because [bp+4] is being used as the first parameter to SendMessage(),

it must be an HWND of some sort.

Here is the muster for SendMessage(): LRESULT SendMessage(HWND hWnd,

UINT uMsg, WPARAM wMsgParam1, LPARAM lMsgParam2), where hWnd

identifies the Window receiving the message, uMsg identifies the

message being sent, wMsgParam1 & lMsgParam2 contain 16 bits and 32

bits of message-specific information.

Finally, we don't see anything being moved into AX or DX near the end

of the function, so it looks as if this function has no return value:

void near pascal some_func(HWND hwnd)

Let's look once more at it... the function starts off with two nested

calls to SendMessage (using the message numbers 41Ah and 409h). These

numbers are greater than 400h, they must therefore be WM_USER+XX

values. Windows controls such as edit, list and combo boxes all use

WM_USER+XX notification codes.

The only appropriate control in TASKMAN is the list box, so we can

just look at the list of LB_XXX codes in WINDOWS.H. 1Ah is 26 decimal,

therefore 41Ah is WM_USER+26, or LB_GETITEMDATA. Let's see what

Osborne's "Special Purpose API functions" says about it (pag.752):

LB_GETITEMDATA

When sent: To return the value associated with a list-box item.

wParam: Contains the index to the item in question

lParam: Not used, must be 0

Returns: The 32-bit value associated with the item

Similarly, 409h is WM_USER+9, which in the case of a list box means

LB_GETCURSEL. We saw earlier that TASKMAN uses LB_SETITEMDATA to store

each window title's associated HWND. LB_GETITEMDATA will now retrive

this hwnd:

hwnd = SendMessage(listbox, LB_GETITEMDATA,

SendMessage(listbox, LB_GETCURSEL, 0, 0), 0);

Notice that now we are caling the parameter to some_func() a listbox,

and that the return value from LB_GETITEMDATA is an HWND.

How would we know it's an hwnd without our references? We can see the

LB_GETITEMDATA return value (in DI) immediatly being passed to

IsWindow() at line 1.0039:

; IsWindow(hwnd = SendMessage(...));

1.0031 9AEF010000 call far ptr SENDMESSAGE

1.0036 8BF8 mov di, ax

1.0038 57 push di

1.0039 9A4C000000 call far ptr ISWINDOW

Next, the hwnd is passed to GetLastActivePopup(), and the HWND that

GetLastActivePopup() returns is then checked with IsWindow()...

IsWindow() returns non-zero if the specified hWnd is valid, and zero

if it is invalid:

; IsWindow(hwndPopup = GetLastActivePopup(hwnd));

1.0042 57 push di

1.0043 9AFFFF0000 call USER.GETLASTACTIVEPOPUP

1.0048 8BF0 mov si, ax ; save hwndPopup in SI

1.004A 56 push si

1.004B 9AA4020000 call USER.ISWINDOW

Next, hwndPopup (in SI) is passed to GetWindowLong(), to get

informations about this window. Here is time to look at WINDOWS.H to

figure out what 0FFF0h at line 1.055 and 800h at line 1.005C are

supposed to mean:

; GetWindowLong(hwndPopup, GWL_STYLE) & WS_DISABLED

1.0054 56 push si ;hwndPopup

1.0055 6AF0 push GWL 5STYLE ;0FFF0h = -16

1.0057 9ACD020000 call USER.GETWINDOWLONG

1.005C F7C20008 test dx, 0800 ;DX:AX= 800:0= WS_DISABLED

Finally, as the whole point of this exercise, assuming this checked

window passes all its tests, its last active popup is switched to:

; SwitchToRhisWindow(hwndPopup, TRUE)

1.0062 56 push si ;hwndPopup

1.0063 6A01 push 0001

1.0065 9AFFFF0000 call USER.SWITCHTOTHISWINDOW

It's here that all possible questions START: SwitchToThisWindow is not

documented... therefore we do not know the purpose of its second

parameter, apparently a BOOL. We cannot even tell why

SwitchToThisWindow() is being used... when SetActiveWindow(),

SetFocus() or BringWindowToTop() might do the trick. And why is the

last active popup and not the window switched to?

But let's resume for now our unearthed mysterious function, that will

switch to the window selected in the Task List if the window meets all

the function's many preconditions:

void MaybeSwitchToSelectedWindow(HWND listbox)

{

HWND hwnd, hwndPopup;

// first figure out wich window was selected in the Task List

if (IsWindow(hwnd = SendMessage(listbox, LB_GETITEMDATA,

SendMessage(listbox, LB_GETCURSEL, 0, 0), 0)))

{

if (IsWindow(hwndPopup = GetLastActivePopup(hwnd)))

{

if (! (GetWindowLong(hwndPopup, GWL_STYLE) & WS_DISABLED))

{

SwitchToThisWindow(hwndPopup, TRUE);

return;

}

}

MessageBeep(0); //Still here... error!

}

Now we have a good idea of what TASKMAN does (it sure took a long time

to understand those 3K bytes of code!). In the next lessons we'll use

what we have learned to crack together some common Windows programs.

(->lesson 3)

FraVia